View Code? Open in Web Editor
NEW
License: Apache License 2.0
Dockerfile 1.76%
Go 33.32%
Shell 2.16%
TypeScript 46.78%
HTML 14.93%
CSS 1.05%
reval's Introduction
checkout repository
create folder tmp
place files Llama-2-13b-chat-hf_responses.json and vicuna-33b-v1.3_responses.json into tmp folder
rename the files to import_Llama-2-13b-chat-hf_responses.json and import_vicuna-33b-v1.3_responses.json
run docker compose up
open http://localhost:8080
reval's People
Contributors
Stargazers
Watchers
reval's Issues
The rating score seems not to work as intended.
At any given time, all users see the same conversation. This results in every conversation being evaluated multiple times, instead of multiple users evaluating different conversations.
Create CSV export for evaluations. One record per response:
Columns:
request.Id
response.Id
for each user one column with user.name
Rows:
evaluation.evaluation_result for each user
Llama-2-13b-chat-hf_sampleSet2_responses_rated_by_gpt-3.5-turbo.json contains the new format